All Environments¶

MOMAland includes environments taken from the MO/MARL literature, as well as multi-objective versions of environments from PettingZoo. More information are available in the TODO MOMAland paper.

Env	Cooperative/Adversarial	Obs/Action spaces	Objectives	Description
`mo-beach-v0`	Any	Discrete / Discrete	`[occupation, mixture]`	Taken from Mannion_2018. MO-Beach is a game with two objectives, reflecting the enjoyment of tourists (agents) on their respective beach sections in terms of crowdedness and diversity of attendees. Each beach section is characterised by a capacity and each agent is characterised by a type.
`mo-item-gathering-v0`	Adversarial	Discrete / Discrete	`[#objects]` (configurable)	Adapted from Kallstrom_2019, is a multi-agent grid world, containing items of different colours. Each colour represents a different objective and the goal of the agents is to collect as many objects as possible.
`mo-gem-mining-v0`	Cooperative	- / Discrete	`[#gems]` (configurable)	MO version of Gem Mining Bargiacchi_2018. Agents go to different mines to extract different gems (objectives). There are restrictions on which mines can be reached for each agent. Agents also influence each other’s producitivity.
`mo-route_choice-v0`	Adversarial	- / Discrete	`[travel time, cost]`	MO-RouteChoice is a multi-objective extension of the route choice problem Thomasini_2023, where a number of self-interested drivers (agents) must navigate a road network.
`mo-pistonball-v0`	Cooperative	Continuous / Any	`[agent_#n_reward]` (configurable)	An MO version of PZ’s Pistonball where the reward of each agent is kept separate.
`mo-multiwalker-stability-v0`	Cooperative	Continuous / Continuous	`[progress right, package stability]`	A MO version of PZ’s MultiWalker introduced in Gupta_2017, where the agents also seek to keep the package steady.
`catch-v0`	Cooperative	Continuous / Continuous	`[distance_target, distance_other_drones]`	Agents must corner and catch a target drone while maintaining distance between themselves.
`escort-v0`	Cooperative	Continuous / Continuous	`[distance_target, distance_other_drones]`	Agents must circle around a mobile target drone and escort it to its destination without breaking formation while maintaining distance between themselves.
`surround-v0`	Cooperative	Continuous / Continuous	`[distance_target, distance_other_drones]`	Agents must surround a fixed target point while maintaining distance between themselves.
`mo-breakthrough-v0`	Adversarial	Discrete / Discrete	`[win, fast win, capturing opponent's pieces, avoiding capture]`	Multi-objective version of the two-player, turn-based, board game Breakthrough.
`mo-connect4-v0`	Adversarial	Discrete / Discrete	`[win, fast win, [column #n]]`	MO version of Connect 4. Additional objectives are fast win and optionally one objective per column.
`mo-ingenious-v0`	Any	Discrete / Discrete	`[#colors]` (configurable)	MO adaptation of the zero-sum, turn-based board game Ingenious. The game’s original rules support 2-4 players collecting scores in multiple colors (objectives), with the goal of winning by maximizing the minimum score over all colors. In MO-Ingenious, we leave the utility wrapper up to the users and only return the vector of scores in each colour objective.
`mo-same-game-v0`	Any	Discrete / Discrete	`[colors_n]` (configurable)	MO-SameGame is a multi-objective, multi-agent variant of the single-player, single-objective turn-based puzzle game called SameGame Baier_2015. The original single-player, single-objective SameGame rewards the player with \(n^2\) points for removing any group of \(n\) tiles. MO-SameGame can extend this in two ways. Agents can either only get points for their own actions, leading to competition between them, or all rewards can be shared in ``team reward’’ mode. Additionally, points for every colour can be counted as separate objectives, allowing for different trade-offs between colours, or they can be accumulated in a single objective like in the default game variant, essentially providing a single-objective wrapper for the game.